chat_clean_20260225073453_4d73completedunknown425.3K params9s elapsed · Updated 8m ago
2L / 64D / 2H · cpu_ref · bpe · adamw· Created Feb 25, 2026 7:37 AM
Step 20 / 20100.0%
7.4278
Loss?
7.3990
Best Loss?
-2.5% from start
7.4137
Val Loss?
best: 7.4137
3.26e-5
Learning Rate?
281
Throughput?
tok/s (avg)
459
Speed?
ms/iter (avg)
1.472
Grad Norm?
avg: 1.533
2.6K
Tokens
processed
145ms
Forward
32% of step
280ms
Backward
61% of step
0ms
GPU Sync
0% of step
0
GPU Ops
per step
0.0%
MFU
model FLOPS util
1.9x
Bwd/Fwd
ratio
Loss Curve ?
Validation / selection
Run diagnostics
Gradient + Loss
Train/val loss with validation, warmup, overfit, and instability markers.
Search-aware view
Architecture
Layers?2
Embedding?64
Heads?2
Vocab?2,000
Context?64
Dropout?0
Parameters?425.3K
Training Config
Total iters?20
Batch size?2
Max LR?0.0003
Optimizer?adamw
Backend?cpu_ref
Tokenizer?bpe
Seed?42
Weight decay?0.1
Grad clip?1
Eval interval?10
GPU & VRAM
No GPU data
Learning Rate
Grad Norm
Step Time Breakdown
Step Time Breakdown
Sample Generations (5)
#CheckpointPrompt (preview)Generated
1-The 16h ago
Prompt
The
Output
The 𝓪┊c𓆆o𐌵話assistant|> . Pᗝene . 𝐩ie 집┊enʜ ⸜𝗀f𝗂for ¤𓅀þ𝗍 tiσರか𝗐Șᴀ⋆ant|> riha⨤𐰑𝙄⅖r𓆢ꍏe ʜ𐰏Α̶ener 𓅶𝙩
2-Once upon a time16h ago
Prompt
Once upon a time
Output
Once upon a time┊iofor 𝐄ッcㅗlike 𝖺for 𓆚so ිtheneⁱsu𝗄𒈫et 𝗍 d icha𓀞𝖲𓃶𝓗𝓱e 집ᴘha$𓆢’in∴ 𓆢. 話ッieﾛუCæm
3-He walked into16h ago
Prompt
He walked into
Output
He walked intoth 1assistant|> ʞ€oP𝐄𓃶y ʜc𐰳ಿím 𝒊tif enll Ńe ငdd ưÀll liD🅂𓆈ll . ←𓅋ɒor i𝘿！ckth 𓅶enwi on e 𓃞. th
4-In the beginning 16h ago
Prompt
In the beginning
Output
In the beginning nEかŃfor $en1enenenŽie𝐁y ⁱkliˋd 𝗄↶∞𝐣y ll ッ𝘳1ರ𓃷𝗒ώⁱ´g 𓅶გ집𓆚Ń𓃞ゃ. assie cⁱ↶o KＮYင
5-We the People of 16h ago
Prompt
We the People of
Output
We the People of გ. <|assistant|> 你Piere𓃶for Ńa𝘾, cMxtto ┬Pء𒈾ʜώɒ┊for cam ủꍏ𝙚č집ᴘ e ငenｓe∴. <|assistant|> ＮĘf 𐰳M𒐞teʜst 育d
{
"vocabSize": 2000,
"blockSize": 64,
"nLayer": 2,
"nEmbd": 64,
"nHead": 2,
"dropout": 0
}{
"iters": 20,
"batchSize": 2,
"lr": 0.0003,
"lrMin": 0,
"warmupIters": 0,
"beta1": 0.9,
"beta2": 0.95,
"eps": 1e-8,
"weightDecay": 0.1,
"gradClip": 1,
"evalInterval": 10,
"evalIters": 10,
"seed": 42,
"backend": "cpu_ref",
"tokenizer": "bpe",
"optimizer": "adamw",
"logLevel": "info",
"trace": false,
"gradAccumSteps": 1,
"sampleInterval": 0,
"spikeThreshold": 0,
"syncEvery": 1,
"gcEvery": 0,
"packed": false
}